OPTIMAL SUPPORTS FOR LINEAR PREDICTIVE MODELS

The problem of finding the optimal set of causal pixels (support) for use in linear predictive models is addressed. After presenting counter examples to popular intuitions about supports, a general result relati ng the distortion incurred with a small support to optimal coefficient s of a larger support is derived. A geometrical interpretation is prov ided. Two algorithms that optimally increase/decrease support sizes by one at each step are presented, Experimental results illustrate the s ignificant gains realized by the algorithms compared with commonly use d supports.